Update README.md

README.md

@@ -23,7 +23,72 @@ This llama model was trained 2x faster with [Unsloth](https://github.com/unsloth
 
 # How to conduct inference
 
-`import XXXX`
+```
+from unsloth import FastLanguageModel
+from peft import PeftModel
+import torch
+import json
+from tqdm import tqdm
+import re
+
+# Base model id and LoRA adapter ID
+base_model_id = "llm-jp/llm-jp-3-13b"
+adapter_id = "Rumi/llm-jp_SFT_rn_2024-12-14_06"
+
+# Log in with your Hugging Face token
+HF_TOKEN = "hogehoge"
+from huggingface_hub import login
+login(HF_TOKEN)
+
+# Download the original model
+dtype = None
+load_in_4bit = True
+
+base_model, tokenizer = FastLanguageModel.from_pretrained(
+    model_name=base_model_id,
+    dtype=dtype,
+    load_in_4bit=load_in_4bit,
+    trust_remote_code=True,
+)
+
+# Merge adapter to the base model
+model = PeftModel.from_pretrained(base_model, adapter_id, token = HF_TOKEN)
+
+# Read evaluation dataset
+datasets = []
+with open("./elyza-tasks-100-TV_0.jsonl", "r") as f:
+    item = ""
+    for line in f:
+      line = line.strip()
+      item += line
+      if item.endswith("}"):
+        datasets.append(json.loads(item))
+        item = ""
+
+# Change the format and conduct the evaluation
+FastLanguageModel.for_inference(model)
+
+results = []
+for dt in tqdm(datasets):
+  input = dt["input"]
+
+  prompt = f"""### 指示\n{input}\n### 回答\n"""
+
+  inputs = tokenizer([prompt], return_tensors = "pt").to(model.device)
+
+  outputs = model.generate(**inputs, max_new_tokens = 512, use_cache = True, do_sample=False, repetition_penalty=1.2)
+  prediction = tokenizer.decode(outputs[0], skip_special_tokens=True).split('\n### 回答')[-1]
+
+  results.append({"task_id": dt["task_id"], "input": input, "output": prediction})
+
+# Save result in the jsonl format
+json_file_id = re.sub(".*/", "", adapter_id)
+with open(f"/content/{json_file_id}_output.jsonl", 'w', encoding='utf-8') as f:
+    for result in results:
+        json.dump(result, f, ensure_ascii=False)
+        f.write('\n')
+
+```